DRAFT: Add option to keep oldest triples when dropping duplicates

[alightwing]

We typically create our models over two passes, once to create a model from Excel data, and then a second one to add in labels triples. Since we currently drop the oldest triple where duplicates are found, this means that the labels triples always "win" - however, we usually treat our Excel data as a source of truth so it's these older triples that should be retained.

So, add option to select behaviour when dropping duplicates. Default is the existing keep-newest behaviour as this is appropriate when building the Excel model itself, but allow keep-oldest instead at the user's discretion.

[bz2]

Thanks! This seems fine? I think the least clear part is how safe the newest/oldest assumption with the triple store order is - it may not match intuition if the input arguments and processing steps do not align.